Sources of Uncertainty in Calculating Mortality and Morbidity Attributable to Air Pollution

Abstract

Quantifying the burden of illness and mortality from air pollution exposure relies on statistical estimates and other assumptions that have inherent uncertainties. Through an intensive study in Hamilton, Canada, this study illustrates for policymakers the sensitivity of health effect estimates to a wide range of possible uncertainties. Dose-response relationships were derived based on pooled and averaged estimates published in the scientific literature from 1997 to 2001. These estimates were applied to local air pollution, mortality, and hospital admissions data for the years 1995–1999. The data were adjusted to reflect uncertainties in the current state of knowledge, including (1) baseline pollution, (2) single versus multipollutant effects, (3) local or pooled estimates, and (4) chronic effects. The estimates of mortality ranged from 96 to 374 annual deaths, while admissions ranged from 139 to 607 respiratory and from 479 to 2000 cardiovascular admissions. Chronic fine particle exposure resulted in 232 annual deaths. As conclusions, first, there should be an effort to reach a consensus on reporting scientific findings from air pollution studies using standardized study designs and reporting formats. Second, given the sensitivity of the estimates to underlying assumptions, an immediate need exists for widely accepted burden of illness and mortality estimation conventions. Third, many areas of air pollution research require considerable refinement before complete estimates can be ascribed.

We thank Dr. Tom Abernathy, Central West Health Planning Information Network, for supplying the mortality and morbidity data. Frank Dobroff, Ontario Ministry of the Environment, assisted with the air pollution data. We thank Drs. R. Burnett and S. Cakmak for assistance with programming the random effect models. We acknowledge helpful comments from Dr. Susan Elliott, Sonya Kapusin, Norm Finkelstein, Chris Giovis, Ric Hamilton, and Pat DeLuca. We acknowledge funding from the City of Hamilton, Health Canada, the Canadian Institutes of Health Research, and the Southern California Environmental Health Sciences Center 5P30 ES07048 (funded by the National Institute of Environmental Health Science).

Notes

∗Because the Poisson regression takes a log-linear form, we computed the risk estimates for each criteria pollutant as

where e is the exponential function, β is the regression coefficient estimating the average increase in mortality associated with a unit increase in pollution, and is the average of the air pollutant.